Pentagon slaps anthropic with 'supply-risk' tag, freezing federal contracts
The Pentagon just told a federal judge to keep Anthropic on ice. In a terse filing Tuesday, Department of Defense lawyers argued the ai unicorn’s refusal to let the military use its models for “any lawful purpose” makes it a security wild card, not a partner.
The contract clause that started the freeze
Last month, the Trump administration ordered agencies to stop integrating Claude into logistics, intel and battlefield-simulation tools. The memo arrived after Anthropic balked at amending its usage policy, which bans autonomous-weapons work and dragnet surveillance of U.S. citizens. Defense officials claim that reservation gives a private lab veto power over missions. Anthropic calls it a principled red line.
Now the company is suing for access to what it says is a $4-billion federal pipeline. The government counters that procurement teams can shop elsewhere and labels Anthropic’s First-Amendment complaint “a commercial gripe dressed up as free speech.”
Inside the classified briefing
According to sealed excerpts leaked to reporters, Pentagon cyber officers warned that Anthropic’s cloud-tethered update cycle could let engineers alter model weights mid-operation, “effectively rewriting the software while a convoy is on the move.” The nightmare scenario: a silent downgrade that mislabels a hospital as a weapons depot or misses a troop movement.
Anthropic’s reply is blunt: the same code that powers customer-support bots for Marriott and Notion isn’t suddenly a national-weaponization kit. Yet the firm’s own safety paper from October concedes that “runtime steering remains technically feasible,” a concession the Justice Department is waving like a red flag.
What’s really at stake on march 24
When both sides meet in San Francisco for the injunction hearing, the judge will weigh two competing futures: a market where startups dictate military ethics, or one where Washington blacklists any lab that won’t sign a blank check. Anthropic says a loss would crater U.S. leadership in responsible ai. The Pentagon shrugs: there are 1,200 other vendors in the queue.
Lo que nadie cuenta es que the same day the filing landed, Palantir closed a $3.4-billion defense expansion, its stock popping 11 % after-hours. The message from investors is unmistakable: whatever Claude can’t do, someone else will—at a premium.
The gavel falls in eleven days. By lunch, we’ll know if Anthropic returns to the federal buffet or remains the smartest kid barred from the cafeteria.